Table 3 AUC (adjusted) values for differentiation of CD vs. non-IBD, UC vs. non-IBD, and CD vs. UC using a GB classifier with 5-fold cross-validation under MAR, MCAR, and MNAR mechanisms.

Percentage	MAR		MCAR		MNAR
CD vs. non-IBD
5	II (BR)	0.859	II (BR)	0.861	II (BR)	0.855
10	II (BR)	0.849	II (BR)	0.854	II (BR)	0.843
15	II (BR)	0.841	II (BR)	0.848	II (BR)	0.832
20	II (BR)	0.834	II (BR)	0.840	II (BR)	0.825
25	II (BR)	0.827	II (BR)	0.833	II (BR)	0.817
30	II (BR)	0.821	II (BR)	0.825	II (BR)	0.814
35	II (BR)	0.813	AE	0.815	AE	0.806
40	II (BR)	0.804	AE	0.806	AE	0.804
Original data	0.866
UC vs. non-IBD
5	AE	0.805	AE	0.804	II (RF)	0.804
10	AE	0.795	AE	0.795	II (RF)	0.782
15	AE	0.785	AE	0.785	II (RF)	0.770
20	AE	0.775	AE	0.775	II (RF)	0.759
25	AE	0.741	AE	0.763	AE	0.749
30	AE	0.719	AE	0.751	AE	0.738
35	AE	0.699	AE	0.738	AE	0.731
40	II (BR)	0.684	AE	0.724	AE	0.721
Original data	0.816
CD vs. UC
5	II (BR)	0.821	II (BR)	0.821	II (RF)	0.821
10	II (BR)	0.819	II (BR)	0.818	II (BR)	0.813
15	II (BR)	0.816	II (BR)	0.815	II (BR)	0.807
20	II (BR)	0.814	II (BR)	0.811	II (BR)	0.803
25	II (BR)	0.812	II (BR)	0.806	II (BR)	0.798
30	II (BR)	0.809	II (BR)	0.800	II (BR)	0.793
35	II (BR)	0.806	AE	0.793	II (BR)	0.788
40	II (BR)	0.801	AE	0.787	AE	0.784
Original data	0.823

Iterative imputers (especially II(BR)) consistently maintain the highest classification performance in CD-related tasks, while AE becomes competitive at higher missingness levels. In UC vs. non-IBD, AE generally outperforms other methods across most missingness degrees, whereas II(RF) shows an advantage in MNAR at low missingness.

Quick links

Search